I recently made an unusual audio processing challenge where the goal is to recover the English text that a user is typing, just by analysing the recorded sound of their keyboard [0]. Not sure how hard it is. I am able to solve it with my own audio processing tools. If interested - give it a try.